Ultralytics
imx500: Running network.rpk on imx500 leads to: ValueError: tensor info length 260 does not match expected size 708 #17762
Replies: 5 comments 25 replies
-
|
👋 Hello @JochenSchmitz, thank you for reaching out about running your network on the imx500. It's great to see you're diving into Ultralytics 🚀! We recommend checking out our Docs, where you might find helpful information and examples for typical setups. If this is indeed a 🐛 Bug Report, please help us diagnose the issue more effectively by providing a minimum reproducible example. This will allow us to better understand the scenario and identify any potential problems. For questions related to custom training, please share as much relevant information as possible, such as dataset examples and any training logs. You should also ensure you're following our Tips for Best Training Results. Join the vibrant Ultralytics community for more interactive discussions! Whether it’s through Discord for real-time chats 🎧, Discourse for more detailed discussions, or our Subreddit for sharing insights, there are plenty of places to connect with fellow users. UpgradePlease make sure you’re using the latest version of the pip install -U ultralyticsEnvironmentsYOLO can operate in a range of verified environments. Here's how you can get started with each:
StatusVerify our system's health with this CI badge: This is an automated response 😊, and rest assured an Ultralytics engineer will follow up to assist you further! |
Beta Was this translation helpful? Give feedback.
-
|
I have exactly the same problem! I was able to get my rtk file for my custom YOLO model. Can someone help? |
Beta Was this translation helpful? Give feedback.
-
|
This issue seems related to the firmware or metadata parsing on the IMX500 and not specifically tied to YOLO or Ultralytics. For YOLO model-specific issues, ensure the model is correctly formatted and validated. For IMX500-related help, I recommend consulting the camera or firmware documentation or reaching out to their support community. |
Beta Was this translation helpful? Give feedback.
-
|
@JochenSchmitz hi Jochen, the error suggests a mismatch in tensor information length, which could be due to an issue with the exported |
Beta Was this translation helpful? Give feedback.
-
|
@edouardrolland thank you for the update! The issue may stem from the IMX500 firmware or Python script compatibility. Please confirm that the model input/output dimensions are correctly aligned with your setup and ensure the IMX500-specific configuration matches the exported model. Revisiting Sony IMX500 integration documentation might help pinpoint discrepancies. Let us know if you need further assistance! |
Beta Was this translation helpful? Give feedback.
-
|
Hi @glenn-jocher Thank you so much for your answer. Below are the different steps I followed and the associated file.
Side question: Will I get better performance with my custom validation set or by using the COCO dataset? Here you can find the output of the compression: compression output Let me know if you see something wrong here! Thank you! |
Beta Was this translation helpful? Give feedback.
-
|
@glenn-jocher As a reference, you can find the full raspberry pi output after the model export: |
Beta Was this translation helpful? Give feedback.
-
|
@glenn-jocher As mentioned below, I was able to resolve the issue by switching the camera branch. However, if you have any comments on my compressed model, I'm all ears :) |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the update! If your compressed model performs as expected on validation and deployment, it seems you have successfully addressed the key requirements. For additional feedback, I recommend reviewing the quantization logs for warnings and ensuring mAP performance aligns with expectations. Let us know if further assistance is needed! |
Beta Was this translation helpful? Give feedback.
-
|
I am facing the same issue. |
Beta Was this translation helpful? Give feedback.
-
|
@mzahana this error typically indicates a mismatch between the model export configuration and the IMX500 firmware requirements. Please ensure that your model was exported with the correct configurations, including INT8 quantization. Verify your setup aligns with the steps outlined in the documentation, and update your firmware or tools if needed: https://docs.ultralytics.com/integrations/sony-imx500/#package-model-and-deploy-to-ai-camera. |
Beta Was this translation helpful? Give feedback.
-
|
@glenn-jocher I did all that. Int8=true is the default value when exporting a model. Could it be that special information has to be provided when training the model? For example, imgsz or other? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @JochenSchmitz, do you mind sharing your exporting terminal output? It would be really useful if we can compare them! Thank you! |
Beta Was this translation helpful? Give feedback.
-
|
It is exactly the same as yours. I posted it on Top of this discussion |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for confirming. If your export logs and configurations are identical, the issue might stem from compatibility between your model and the IMX500 firmware version. Please ensure your firmware is fully updated, and double-check all export parameters, such as input size ( |
Beta Was this translation helpful? Give feedback.
-
|
I followed the solution mentioned here I checked out Then, I ran the python object detection script. The model was uploaded successfully and it was running fine one the camera. |
Beta Was this translation helpful? Give feedback.
-
|
Hurray. Now it works. Many thanks to all of you. :-) The only Problem now is that the bboxes are displaced because the --bbox-order parameter is not accepted. But i'll look into that now |
Beta Was this translation helpful? Give feedback.
-
|
I have the same issue, if you find sthg let me know! |
Beta Was this translation helpful? Give feedback.
-
|
@edouardrolland please share more details about your issue, including any error messages or steps to reproduce it, so I can assist you effectively. |
Beta Was this translation helpful? Give feedback.
-
|
@edouardrolland I will do more work with it tomorrow. It seems, that there are different demo scripts. When I have the solution, I will post it here. Now its Python. Not Hardware ;-) |
Beta Was this translation helpful? Give feedback.
-
|
detect.txt |
Beta Was this translation helpful? Give feedback.
-
Hello, i finally managed to come to the Point where i have the network.rpk File. Starting the demo now gives me the result:
imm@JSPI10:~/Projekte/aiCam/picamera2/examples/imx500 $ python imx500_object_detection_demo.py --model ~/Projekte/aiCam/best_imx_model/network.rpk
[0:27:41.230457853] [3907] INFO Camera camera_manager.cpp:325 libcamera v0.3.2+27-7330f29b
[0:27:41.237921372] [3911] INFO RPI pisp.cpp:695 libpisp version v1.0.7 28196ed6edcf 29-08-2024 (16:33:32)
[0:27:41.238687766] [3911] WARN CameraSensorProperties camera_sensor_properties.cpp:305 No static properties available for 'imx500'
[0:27:41.238697544] [3911] WARN CameraSensorProperties camera_sensor_properties.cpp:307 Please consider updating the camera sensor properties database
[0:27:41.248370265] [3911] INFO RPI pisp.cpp:1154 Registered camera /base/axi/pcie@120000/rp1/i2c@88000/imx500@1a to CFE device /dev/media1 and ISP device /dev/media0 using PiSP variant BCM2712_C0
NOTE: Loading network firmware onto the IMX500 can take several minutes, please do not close down the application.
[0:27:41.291403903] [3907] INFO Camera camera.cpp:1197 configuring streams: (0) 640x480-XBGR8888 (1) 2028x1520-RGGB_PISP_COMP1
Network Firmware Upload: 0.00bytes [00:00, ?bytes/s][0:27:41.292237317] [3911] INFO RPI pisp.cpp:1450 Sensor: /base/axi/pcie@120000/rp1/i2c@88000/imx500@1a - Selected sensor format: 2028x1520-SRGGB10_1X10 - Selected CFE format: 2028x1520-PC1R
Network Firmware Upload: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2.89M/2.89M [00:12<00:00, 242kbytes/s]
Traceback (most recent call last):
File "/home/imm/Projekte/aiCam/picamera2/examples/imx500/imx500_object_detection_demo.py", line 179, in
last_results = parse_detections(picam2.capture_metadata())
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/imm/Projekte/aiCam/picamera2/examples/imx500/imx500_object_detection_demo.py", line 33, in parse_detections
np_outputs = imx500.get_outputs(metadata, add_batch=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/imm/Projekte/aiCam/picamera2/picamera2/devices/imx500/imx500.py", line 474, in get_outputs
output_shapes = self.get_output_shapes(metadata)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/imm/Projekte/aiCam/picamera2/picamera2/devices/imx500/imx500.py", line 491, in get_output_shapes
output_tensor_info = self.__get_output_tensor_info(output_tensor_info)['info']
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/imm/Projekte/aiCam/picamera2/picamera2/devices/imx500/imx500.py", line 548, in __get_output_tensor_info
raise ValueError(f'tensor info length {len(tensor_info)} does not match expected size {size}')
ValueError: tensor info length 260 does not match expected size 708
imm@JSPI10:~/Projekte/aiCam/picamera2/examples/imx500 $
Can anyone help, please?
Thanks
Jochen
Beta Was this translation helpful? Give feedback.
All reactions